You are on page 1of 17

2021 Spring

Model Predictive Control for Path Tracking of


Autonomous Vehicle

2021.7.06

Bilal Sadiq
Department of Robotics Engineering
Daegu Gyeongbuk Institute of Science and Technology (DGIST)

Bilal Sadiq – bilalsadiq@dgist.ac.kr 1


Today’s Presentation Agenda

1. Constraints and Optimization

2. Comparison and Simulations

Bilal Sadiq – bilalsadiq@dgist.ac.kr 2


1. Modified the Constraint on Control signal

Bilal Sadiq – bilalsadiq@dgist.ac.kr 3


Bilal Sadiq – bilalsadiq@dgist.ac.kr 4
Bilal Sadiq – bilalsadiq@dgist.ac.kr 5
1. Constraint and Optimization

1. Constraints
 Imposes some restrictions on the model
 In Model predictive control System ,usually three types of constraints

  constraint on the Control variable


constraint on the rate of change of control variable

 Constraint on the amplitude of the output

 Formulation of Constraint in MPC Terminology


 To apply the “ Inequallity constraints” to a system for quadratic programming they
should be property assemble as shown in next slide

Bilal Sadiq – bilalsadiq@dgist.ac.kr 6


1. Constraint and Optimization
1.C Constraints
 Formulation of Constraint in MPC Terminology
 A general form of representation is
𝑀
  Χ≤𝛾
 So,the above inequality can be written as
 

 − 𝐼 − Δ 𝑈 𝑚𝑖𝑛
[𝐼] ∆𝑈 ≤
Δ𝑈[ 𝑚 𝑎𝑥 ]
  Formulation of Constraint in MPC Terminology for Optimal
dU=quadprog(H,G,[],[],[],[],[],[],[],options)

Bilal Sadiq – bilalsadiq@dgist.ac.kr 7


2. Comparison and Simulations

 The path under study is the double lane change maneuvering

Bilal Sadiq – bilalsadiq@dgist.ac.kr 8


2. Comparison and Simulations
2.a Early Results
 𝑟 𝑤 =0 𝑁 𝑝 =30 𝑁 𝑐 =5

Bilal Sadiq – bilalsadiq@dgist.ac.kr 9


2.Simulation Results
2.a New Results
I. Unconstrained with the following MPC Architecture 𝑥  ( 𝑘 }=[ Δ 𝑥 ( 𝑘 ) 𝑦 ( 𝑘 ) ] ′
𝑇 −1 𝑇
𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝐼𝑛𝑝𝑢𝑡  
∆ 𝑈 = ( Φ Φ+ 𝑅 ) Φ (𝑅𝑠 − 𝐹 ∗ 𝑥 ( 𝑘 𝑖 )
)
 
𝑇𝑢𝑛𝑖𝑛𝑔
  𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 𝑎𝑟𝑒
A.
i.  = 0.005
ii. = 15  No Change at all in the responses
iii. =10
iv. =0.05

B.
i.  = 5  Significant Changes Occurs (Results on the next slides)
ii. = 15
iii. =10
iv. =0.05

Bilal Sadiq – bilalsadiq@dgist.ac.kr 10


2.Simulation Results
2.a New Results
I. Unconstrained

 Highly dependent on the control Weight


rather than the MPC parameters.
 Deterioration occurs in the response

Bilal Sadiq – bilalsadiq@dgist.ac.kr 11


2.Simulation Results
 Δ 𝛿
II. Constrained with Same MPC Architecture ≤ Δ 𝛿 𝑓 ≤ Δ 𝛿 𝑓 ,𝑚𝑎𝑥 
𝑓 ,𝑚𝑖𝑛 -
 𝛿 𝑓 , 𝑚𝑖𝑛 ≤ 𝛿 𝑓 ≤ 𝛿 𝑓 , 𝑚𝑎𝑥   -
𝐶𝑜𝑛𝑡𝑟𝑜𝑙
  𝐼𝑛𝑝𝑢𝑡 dU=quadprog(H,G,[],[],[],[],[],[],[],options)
𝑢𝑛𝑖𝑛𝑔 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 𝑎𝑟𝑒
 
R= 1
i. = 15
ii. =10
iii. =0.05

Bilal Sadiq – bilalsadiq@dgist.ac.kr 12


2.Simulation Results
II. Constrained with Same MPC Architecture

 
R= 10
i. = 15
ii. =10
iii. =0.05

Bilal Sadiq – bilalsadiq@dgist.ac.kr 13


2.Simulation Results
II. Constrained with Same MPC Architecture

 
R= 2e3
i. = 15
ii. =5
iii. =0.05

Bilal Sadiq – bilalsadiq@dgist.ac.kr 14


2.Simulation Results

 To claim that our results are “somehow”


close to what we are expecting

Reference
Path Tracking of Autonomous Adaptive Mode
predictive control by Fen Lin et.al

Bilal Sadiq – bilalsadiq@dgist.ac.kr 15


3. Conclusion

IN Next Meeting

 The focus will be on adding the Road Coefficient Friction as Constraint which is one of
the most import and physical phenomena for a practical implementation .

 Try to smoothen the responses further.

Bilal Sadiq – bilalsadiq@dgist.ac.kr 16


Thank you

Bilal Sadiq – bilalsadiq@dgist.ac.kr 17

You might also like